Currently we read a submission, parse the json, validate it, serialise back to a string, send the data to messybrainz, which parses, processes, and sends a result back, then listenbrainz parses the msb response, alters it, serialises it, and sends to kafka.

If we move validation to a special endpoint in messybrainz then we can remove an entire deserialise/serialise cycle which should use less cpu.

If messybrainz decides that the data is invalid it should raise http400, which listenbrainz can catch and resend back to the client.